Chukwujekwu, CKBello-Salau, HMu’azu, MBAliyu, HAOnumanyi, Adeiza JIbrahim, OM2026-08-312026-08-312026-06http://hdl.handle.net/10204/14921The application of machine learning in medical diagnostics has significantly advanced the early detection and classification of diabetes. However, classifiers often suffer from challenges related to high-dimensional data, overfitting, and ineffective hyperparameter tuning, which compromise their predictive performance and generalisation capability. To address these limitations, this study proposes an improved diabetic classification framework using a chaotic Grey Wolf Optimisation (cGWO) algorithm for hyperparameter tuning of a Random Forest (RF) classifier. The cGWO enhances the standard GWO by improving its exploration and exploitation abilities, thus reducing the risk of local optima and premature convergence. Using the BRFSS dataset, the study evaluated the proposed cGWO-RF model and compared its performance against an RF and a standard RF-GWO model. The cGWO-RF outperformed the RF-GWO method, achieving a 0.03% increase in accuracy, 0.51% in sensitivity, 0.31% in precision, 0.51% in F1score, 0.2% in AUC, and 1.46% in average precision-recall because of the balance in exploration and exploitation introduced by the cGWO . These results show the efficacy of the cGWO in optimising RF hyperparameters, leading to more accurate, efficient, and generalisable models for diabetes classification. The study contributes an optimisation strategy that enhances predictive modelling in healthcare and supports more reliable decision-making for early diabetes diagnosis.FulltextenDiabetes classificationGrey Wolf OptimisationMachine learning in healthcareRandom forestHyperparameter tuningConference PresentationN/A